F2000climo 20-yr spinup run does not reach equilibrium for several variables

[Shuai Li]

Hi all,

I’m running a default CESM2 case with F2000climo compset, and I’m trying to assess spin-up stability using Keith Oleson’s script SpinupStability_SP_v10.ncl. I’m surprised that after a 20-year run, several global variables still fail the equilibrium tests unless I loosen thresholds. I focused on three variables: FSH, EFLX_LH_TOT, and GPP. When I set glob_thresh_fsh to 0.06, glob_thresh_lh to 0.06, and glob_thresh_gpp to 0.18, these three variables are in equilibrium. I’d appreciate guidance on whether this is expected for F2000climo and how best to proceed.

case basics:
CESM version: cesm2.1.4
Compset: F2000climo
Run length: 20 years
Output: monthly
case build up code:
./create_newcase --case /glade/u/home/dyang/cases/F_ref --res f09_f09_mg17 --compset F2000climo --project WYOM0195
./xmlchange --subgroup case.run JOB_WALLCLOCK_TIME=06:00:00
./xmlchange INFO_DBUG=3,STOP_N=2,STOP_OPTION=nyears,RESUBMIT=4
./case.setup
vi user_nl_cpl
histaux_a2x3hr = .true.
vi user_nl_clm
hist_nhtfrq = 0
hist_mfilt = 1
vi user_nl_cam
mfilt = 1
nhtfrq = 0
./preview_namelists
./case.build
./case.submit
./xmlchange CONTINUE_RUN=TRUE
./xmlchange STOP_N=2,STOP_OPTION=nyears,RESUBMIT=4
./case.submit

I set subper to 4 and ran the script SpinupStability_SP_v10.ncl, and the result is shown in the attached image.

My questions:
1. Do I need to explicitly specify RUN_STARTDATE as 2000 to avoid a cold start? I didn't specify RUN_STARTDATE when creating the case. Will this cause the case to become a cold start, which takes longer to spin up?
2. Do I need to run a new case with the F2000climo compset and specify the RUN_STARTDATE as 2000-01-01 to shorten the spinup? I read the CTSM manual on spinups, which says that the variables are basically in equilibrium after about 10 years run.

Thanks! Any insights and recommended spin-up strategies would be greatly appreciated.

 

[slevis]

1. I looked at your case's lnd_in file and found that a default 2000 finidat file is specified, so you are not starting cold.
2. If your simulation does not seem to be in equilibrium, your best bet is to run longer.

 

[oleson]

Also, that spinup script is for use with I-case runs where you force CLM with a cycling (repeating) atmosphere. Your atmosphere is interactive and therefore each year will be different, thus causing your plot to be noisy. You might just plot a running average of those variables to examine equilibrium.

 

[Shuai Li]

Also, that spinup script is for use with I-case runs where you force CLM with a cycling (repeating) atmosphere. Your atmosphere is interactive and therefore each year will be different, thus causing your plot to be noisy. You might just plot a running average of those variables to examine equilibrium.

Thanks a lot Keith, this is really a good suggestion. I tried using a running average to determine if the variables I was interested in were in equilibrium, and the results looked good.

 

[Shuai Li]

1. I looked at your case's lnd_in file and found that a default 2000 finidat file is specified, so you are not starting cold.
2. If your simulation does not seem to be in equilibrium, your best bet is to run longer.

Hi slevis, I ran the case for another 20 years (now 40 years in total), but I encountered new problems when trying to perform hybrid run and branch run based on the default run.
We now have a default run case F_ref that uses the F2000climo compset. We then established 3 case run:
1. F_ref_cplhist: A branch run based on the F_ref case, run for one year to obtain histaux_* forcing data to drive the simulation of the I2000Clm50Sp
compset;
2. F_1984: A hybrid run based on the F_ref case, we we modified the fsurfat (use land cover data in 1984) in lnd_in;
3. F_2014: same as the F_1984 case but with land cover data in 2014;
I submitted three jobs, F_1984.o3629383, F_2014.o3629495, and F_ref_cplhist.o3629878, to Derecho. They were killed after running for a short time. Here are the three log files.
I requested help from NCAR, and they said it might be a PIO-related issue and suggested I use CESM v2.1.5 for simulation. Does this mean we need to rerun the F_ref case and then other cases? Can I use the restart file generated by CESM2.1.4 to perform a branch run and a hybrid run in CESM2.1.5? Is there an incompatibility issue with restart files between different CESM sub-versions (2.1.4 → 2.1.5)? What do you recommend?

 

[oleson]

I think you could try using the CESM2.1.4 restart file in CESM2.1.5 simulations. There weren't any science changes in CLM between the two tags (release-clm5.0.035 -> release-clm5.0.037).